Forecasting the value at risk of the crude oil futures market: Do high-frequency data help?

This paper presents the first formal comparison of Value at risk (VaR) forecasting performance across various high-frequency volatility models and conventional benchmarks using daily data in the crude oil futures market. Our analysis reveals the following key findings:(1) High-frequency data significantly enhance the accuracy of VaR forecasts. Specifically, the realized-GARCH (generalized autoregressive conditional heteroskedasticity) model that incorporates 5-s realized bipower variation (BPV) outperforms all other models. (2) Not all realized measures are equally effective for VaR forecasting. The 5-s BPV model consistently outperforms other realized measures in forecasting VaR. (3) The choice of sampling frequency plays a crucial role in the performance of realized measures when forecasting VaR. (4) Many more sophisticated realized measures fail to surpass the simple 5-min realized variance (RV) model in forecasting VaR in the crude oil futures market.